In 15 ds006253 SEEG patients, three 0.2 s HGA bins did not improve offline direction decoding over their mean; bin shape was at chance

We tested a pre-registered question: do time-resolved high-gamma (HGA) bins decode left/right movement direction better than the window mean? We used 15 SEEG patients of OpenNeuro ds006253 and the author 70-150 Hz HGA. We used three 0.2 s bins at 0-0.6 s after mouse-movement onset. The decoder was shrinkage LDA with 5 time-ordered folds. The answer is no. Concatenated bins gave a lower mean AUC than their mean (0.559 vs 0.574; difference -0.015, 95% CI -0.040 to +0.012; 5/15 patients better; sign-flip p = 0.86). The result was the same for 0-1.0 s (4/15; p = 0.94). Bins were above chance (13/15; Holm p = 0.040). Bin shape with the level removed was at chance (AUC 0.509; Holm p = 0.57). Thus, the direction information is in the HGA level, not in its time course at this scale. For a linear SEEG BCI direction decoder, a window mean is a sufficient feature. Limits: offline, one session, n = 15, 0.2 s bins and LDA only.

stereo-electroencephalography · high-gamma activity · movement direction · brain-computer interface · linear discriminant analysis · time-resolved features · OpenNeuro ds006253